A class of machine learning models that learn the underlying probability distribution of training data and can sample novel instances from that distribution. Architectures include generative adversarial networks, variational autoencoders, diffusion models, normalising flows, and autoregressive transformers, covering domains such as text, image, audio, and code generation.
Semantic Classification
Content
Generative Model — content pending enrichment.
Current Landscape (2026)
- The frontier has consolidated around a handful of large multimodal foundation models: OpenAI’s GPT-5 (August 2025) succeeding the GPT-4o line, Google DeepMind’s Gemini 2.5 (2025), Anthropic’s Claude Sonnet 4 and Opus 4 (2025), and Meta’s openly available Llama 3 series, shifting the field from text-only chat toward natively multimodal reasoning systems.
- Generative video reached production quality with OpenAI’s Sora and Google’s Veo, extending high-fidelity diffusion generation from still images into temporally coherent clips and accelerating debate over synthetic media provenance.
- Reasoning-oriented models (OpenAI’s o-series, DeepSeek-R1) mainstreamed inference-time “chain-of-thought” compute, where quality scales with test-time reasoning tokens rather than solely with pre-training size.
- Regulation matured: the EU AI Act entered into force on 1 August 2024, with obligations for general-purpose AI (GPAI) models — including transparency, copyright and systemic-risk documentation — applying from 2 August 2025.
- Provenance and watermarking standardisation advanced through the C2PA Content Credentials specification, adopted by major model and platform providers to label AI-generated media.
- Key players now span proprietary labs (OpenAI, Google DeepMind, Anthropic, xAI) and an open-weight ecosystem (Meta Llama, Mistral, DeepSeek, Alibaba Qwen), with enterprise adoption driving a rapidly growing generative-AI market.
- Open challenges as of 2026 include hallucination and factual reliability, the escalating energy and compute cost of training and inference, unresolved copyright and training-data litigation, and evaluation/safety for increasingly agentic systems.
References
-
- OpenAI (2025). Introducing GPT-5. https://openai.com/index/introducing-gpt-5/
-
- Google DeepMind (2025). Gemini 2.5: our most intelligent AI model. https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/
-
- Anthropic (2025). Introducing Claude 4. https://www.anthropic.com/news/claude-4
-
- European Commission (2024). AI Act enters into force. https://commission.europa.eu/news/ai-act-enters-force-2024-08-01_en
-
- C2PA (2025). Coalition for Content Provenance and Authenticity — Content Credentials. https://c2pa.org/